Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75908.27937948794,
-      "required_pages_per_second_upper_estimate": 0.006579537268838822
+      "required_token_positions_per_second": 75693.67655071105,
+      "required_pages_per_second_upper_estimate": 0.006560936039546338
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9488.534922435992,
-      "required_pages_per_second_upper_estimate": 0.0037851184467991034
+      "required_token_positions_per_second": 9461.709568838882,
+      "required_pages_per_second_upper_estimate": 0.003774417412174438
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18977.069844871985,
-      "required_pages_per_second_upper_estimate": 0.004390011123588869
+      "required_token_positions_per_second": 18923.419137677764,
+      "required_pages_per_second_upper_estimate": 0.004377599976699683
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28465.604767307974,
-      "required_pages_per_second_upper_estimate": 0.0017334059828430235
+      "required_token_positions_per_second": 28385.128706516647,
+      "required_pages_per_second_upper_estimate": 0.0017285054129660914
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28465.604767307974,
-      "required_pages_per_second_upper_estimate": 0.0021913610528246498
+      "required_token_positions_per_second": 28385.128706516647,
+      "required_pages_per_second_upper_estimate": 0.002185165782892939
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28465.604767307974,
-      "required_pages_per_second_upper_estimate": 0.3534475429592358
+      "required_token_positions_per_second": 28385.128706516647,
+      "required_pages_per_second_upper_estimate": 0.3524482996202571
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.6275874297735697,
-    "producer_p95_seconds": 0.29346390000137035,
-    "gpu_update_seconds": 0.9712774496101002,
-    "producer_batches_per_second": 1.5934034885956763,
-    "gpu_batches_per_second": 1.029571931687933,
-    "producer_headroom": 1.5476368766030448,
+    "producer_batch_seconds": 0.06373637995304193,
+    "producer_p95_seconds": 0.21087569300107134,
+    "gpu_update_seconds": 0.9740311656100605,
+    "producer_batches_per_second": 15.689626563930279,
+    "gpu_batches_per_second": 1.0266611945354691,
+    "producer_headroom": 15.282185250051578,
     "sustainable": true,
-    "buffer_seconds": 60.219201875826215,
+    "buffer_seconds": 60.38993226782375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 234605.4285099929,
-    "upload_bytes_per_second": 6302700.702389904,
+    "generation_bytes_per_second": 235726.45410890004,
+    "upload_bytes_per_second": 6271494.639577618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.340863623379136,
+    "estimated_full_shards_per_hour": 50.58140962076426,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.0970937430001868,
-      "bytes": 8464356,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.18106462400101,
       "bytes": 6740941,
@@ -562,6 +556,12 @@
       "bytes": 8487979,
       "pages": 27,
       "shards": 4
+    },
+    {
+      "seconds": 1.0571894279983098,
+      "bytes": 6049663,
+      "pages": 10,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 14800, "loss": 0.6797532290220261, "lr": 0.0003, "grad_norm": 0.135252270166779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2165650987, "tokens_per_second": 149397.83953967682, "data_wait_s": 1.0089001079904847e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 14900, "loss": 0.7111192606389523, "lr": 0.0003, "grad_norm": 0.1485962875691798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2180136936, "tokens_per_second": 148151.89915795487, "data_wait_s": 1.0784999176394194e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 15000, "loss": 0.6725511886179447, "lr": 0.0003, "grad_norm": 0.14628862923077823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2194710862, "tokens_per_second": 127660.50463823703, "data_wait_s": 1.1482999980216846e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 15100, "loss": 0.6722135487943888, "lr": 0.0003, "grad_norm": 0.15125409810689924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2209309355, "tokens_per_second": 116584.98687357442, "data_wait_s": 7.790999006829225e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b543c45c30270db6e07cca963a7b93a2b6e854168da27d9c40da75f2db292900
-size 314895
+oid sha256:821088a2947f780b9f85603178520bc7ba032d5df29ec3eb81f288734cf885e4
+size 316978